Re: Ibuffer: w and B default to buffer at current line

[Eli Zaretskii]

> From: John Wiegley <address@hidden>
> Cc: address@hidden,  address@hidden,  address@hidden
> Date: Sat, 17 Sep 2016 14:35:57 -0700
> 
> > If DWIM is okay in the UI, then functions that behave in support of that UI
> > should also be okay.
> 
> Since this responses surprised me a bit, I'm going to assume that I've
> misunderstood somehow.
> 
> Let me give a hypothetical example: Assume for the sake of argument that
> `count-lines' did not report characters as well, and that someone wished to
> extend the `count-lines' command so that, given a prefix arg, it would count
> characters instead of lines.
> 
> What I think should happen in this case is that a new command be created:
> count-items-in-region, which by default counts lines, but with a prefix
> argument counts characters. This leaves that command open to many new
> behaviors in future, while `count-lines' keeps doing just what it says: count
> lines.

If we have 2 functions, one only for counting lines and another only
for counting characters, then how will we give a user a command that
can do both depending on the prefix argument?

In Emacs, every command is a function, so eventually, we would need to
have a single function which could do both, right?  Therefore, the
only issue here is whether that function should be called count-lines
or something else, right?

Now let's make one step forward, and imagine a command that decides
whether or not to count characters as well as lines based on some
context, whatever that may be, and imagine that such a command could
reliably make that decision.  We will then have a function with such a
DWIMish modus operandi, which is totally okay, and is actually the
_only_ way of having such commands in Emacs -- by writing a function
which implements that DWIM.

IOW, as long as having DWIM in a command is okay, it should also be
okay in a function, because in Emacs every command is a function.

> What I would object to is adding a new argument to `count-lines', called
> `characters-p', that changes the behavior of count-lines to now count
> characters instead (again, remember this is hypothetical, I know that today's
> `count-lines' already counts characters as well).

Then how would you implement a command which could do both, under some
circumstances?

> Just because I want DWIM from my interface, doesn't mean I need to implement
> it as DWIM in my functions.

But there are no user interfaces in Emacs except functions.  And if
you are saying that DWIM can only be present in top-level functions
that have an interactive spec, then (a) I think this is too
restrictive and will cause us to put too much code on the top level,
and (b) we cannot prevent Lisp programs from invoking commands as
functions.

> I believe -- very strongly -- that each function should do one thing
> and one thing well, and this one thing should be documented well. I
> don't like magical functions with lots of alternative behaviors,
> unless it is a command created for the purpose of magically
> dispatching to other functions based on its context of use. Such
> magical interface functions are quite alright in my book; but
> complexifying the behavior of core functions is not.

This all boils down to the definition of "core functions", I guess.
If you want to apply this rule only to core functions, you had better
came up with a very good and precise definition.  I predict that this
would be hard to do, and we will have many incidents of Lisp programs
that try to call non-core DWIM functions and requests to remove a
function from the core to make it more DWIMish.  IOW, I believe this
goal is not reachable in Emacs, not in practice anyway.